Small ball probabilities for Gaussian Markov processes under the \(L_p\)-norm. (Q1879524)

Let \(X(t)\) be a real-valued continuous Gaussian Markov process on the interval \([0,1]\) with mean zero. The author studies small ball probabilities, i.e. the behaviour of \(\log P(\| X(t)\| _p <\varepsilon )\), \(\varepsilon \to 0.\) It is known that if the covariance of \(X(t)\) satisfies \(\sigma (s,t)\neq 0\) for \(0<s\leq t<1\), then it can be represented as \[ \sigma (s,t)= G(\min \{s,t\}) \, H(\max \{s,t\}) \] with \(G>0\), \(H>0\) and \(G/H\) nondecreasing on \((0,1)\). The main theorem concerning small ball probabilities is as follows: Assume that \(G\) and \(H\) are absolutely continuous and \(G/H\) is strictly increasing on \([0,1]\). If \(\sup _{0<t\leq 1} H(t)<\infty \) or \(H(t)\) is nonincreasing in a neighborhood of~\(0\), then \[ \lim _{\varepsilon \to 0} \varepsilon ^2 \log P(\| X(t)\| _p<\varepsilon ) =-\kappa _p \left( \int _0^1 (G'H-H'G)^{p/p+2} \,dt\right)^{(p+2)/p}, \] where \(1\leq p\leq \infty\), \(\kappa_p = 2^{2/p}p(\lambda (p)/p+2)^{(p+2)/p}\) and \[ \lambda (p)= \inf \left \{ \int _{-\infty }^\infty | t| ^p\Phi ^2(t)\, dt + \frac 12 \int _{-\infty }^\infty (\Phi '(t))^2\, dt \mid \Phi \in L_2(-\infty ,\infty ), \| \Phi \| _2=1\right \}. \] This result is obtained from a bit more general statement. The author also presents some applications of the theorem to weighted Brownian motion, stationary Gaussian Markov process and others.